Abstract—Current spectrum cognition methods based on mathematical models and deep learning excel at single tasks with clear inputs and outputs, such as signal detection and classification. However, they generally lack autonomous task decomposition, logical reasoning, and tool coordination capabilities. Consequently, it is difficult for them to directly transform raw signal data into valuable information in complex and dynamic wireless communication environments. Multi-Agent Systems (MAS), due to their autonomy, distributivity, and collaborative nature, show broad application prospects in interdisciplinary fields like communications and artificial intelligence. This paper proposes an MAS-based spectrum cognition method, implemented through a “master-slave” MAS architecture jointly driven by the Qwen3 and MiniCPM4 Large Language Models (LLMs). The system consists of a Master Agent, a Signal Perception Sub-Agent, a Knowledge Retrieval Sub-Agent, and a Cognitive Reasoning Sub-Agent, working together to accomplish collaborative spectrum cognition tasks. In experiments, focusing on Wi-Fi signal analysis, we constructed a signal sample database using Universal Software Radio Peripheral (USRP) platforms. The proposed method automated data analysis, feature extraction, historical data querying, and preliminary signal emitter direction estimation for the database. The results demonstrate that the proposed method achieves autonomous task decomposition, signal sample feature identification, and spectrum cognition for Wi-Fi signal emitters. It highlights the potential of MAS in communication signal analysis and offers a new technical approach for intelligent communication detection, automated signal monitoring, and smart emitter identification.
 
Keywords—multi-agent system, communication signal analyzer, spectrum cognition, large language model


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).